diff --git a/flate/fast_encoder.go b/flate/fast_encoder.go
index 433977767..05c27b822 100644
--- a/flate/fast_encoder.go
+++ b/flate/fast_encoder.go
@@ -7,6 +7,7 @@ package flate
 
 import (
 	"fmt"
+	"math/bits"
 
 	"github.com/klauspost/compress/internal/le"
 )
@@ -150,13 +151,33 @@ func (e *fastGen) matchlen(s, t int32, src []byte) int32 {
 			panic(fmt.Sprint(s, "-", t, "(", s-t, ") > maxMatchLength (", maxMatchOffset, ")"))
 		}
 	}
-	s1 := int(s) + maxMatchLength - 4
-	if s1 > len(src) {
-		s1 = len(src)
+	s1 := int32(s) + maxMatchLength - 4
+	if s1 > int32(len(src)) {
+		s1 = int32(len(src))
 	}
 
+	left := s1 - s
+	n := 0
+	for left >= 8 {
+		diff := le.Load64(src, s) ^ le.Load64(src, t)
+		if diff != 0 {
+			return int32(n + bits.TrailingZeros64(diff)>>3)
+		}
+		s += 8
+		t += 8
+		left -= 8
+	}
+
+	a := src[s:s1]
+	b := src[t:]
+	for i := range a {
+		if a[i] != b[i] {
+			break
+		}
+		n++
+	}
+	return int32(n)
 	// Extend the match to be as long as possible.
-	return int32(matchLen(src[s:s1], src[t:]))
 }
 
 // matchlenLong will return the match length between offsets and t in src.
@@ -177,7 +198,28 @@ func (e *fastGen) matchlenLong(s, t int32, src []byte) int32 {
 		}
 	}
 	// Extend the match to be as long as possible.
-	return int32(matchLen(src[s:], src[t:]))
+	left := int32(len(src)) - s
+	n := int32(0)
+	for left >= 8 {
+		diff := le.Load64(src, s) ^ le.Load64(src, t)
+		if diff != 0 {
+			return n + int32(bits.TrailingZeros64(diff)>>3)
+		}
+		s += 8
+		t += 8
+		n += 8
+		left -= 8
+	}
+
+	a := src[s:]
+	b := src[t:]
+	for i := range a {
+		if a[i] != b[i] {
+			break
+		}
+		n++
+	}
+	return n
 }
 
 // Reset the encoding table.
diff --git a/flate/level1.go b/flate/level1.go
index 61854a352..ec721d859 100644
--- a/flate/level1.go
+++ b/flate/level1.go
@@ -2,9 +2,6 @@ package flate
 
 import (
 	"fmt"
-	"math/bits"
-
-	"github.com/klauspost/compress/internal/le"
 )
 
 // fastGen maintains the table for matches,
@@ -122,32 +119,7 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) {
 
 			// Extend the 4-byte match as long as possible.
 			t := candidate.offset - e.cur
-			var l = int32(4)
-			if false {
-				l = e.matchlenLong(s+4, t+4, src) + 4
-			} else {
-				// inlined:
-				a := src[s:]
-				b := src[t:]
-				left := len(a) - 4
-				for left >= 8 {
-					if diff := le.Load64(a, l) ^ le.Load64(b, l); diff != 0 {
-						l += int32(bits.TrailingZeros64(diff) >> 3)
-						goto endMatch
-					}
-					l += 8
-					left -= 8
-				}
-				a = a[l:]
-				b = b[l:]
-				for i := range a {
-					if a[i] != b[i] {
-						break
-					}
-					l++
-				}
-			endMatch:
-			}
+			l := e.matchlenLong(s+4, t+4, src) + 4
 
 			// Extend backwards
 			for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
diff --git a/flate/matchlen_amd64.go b/flate/matchlen_amd64.go
deleted file mode 100644
index 4bd388584..000000000
--- a/flate/matchlen_amd64.go
+++ /dev/null
@@ -1,16 +0,0 @@
-//go:build amd64 && !appengine && !noasm && gc
-// +build amd64,!appengine,!noasm,gc
-
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-
-package flate
-
-// matchLen returns how many bytes match in a and b
-//
-// It assumes that:
-//
-//	len(a) <= len(b) and len(a) > 0
-//
-//go:noescape
-func matchLen(a []byte, b []byte) int
diff --git a/flate/matchlen_amd64.s b/flate/matchlen_amd64.s
deleted file mode 100644
index 0782b86e3..000000000
--- a/flate/matchlen_amd64.s
+++ /dev/null
@@ -1,66 +0,0 @@
-// Copied from S2 implementation.
-
-//go:build !appengine && !noasm && gc && !noasm
-
-#include "textflag.h"
-
-// func matchLen(a []byte, b []byte) int
-TEXT ·matchLen(SB), NOSPLIT, $0-56
-	MOVQ a_base+0(FP), AX
-	MOVQ b_base+24(FP), CX
-	MOVQ a_len+8(FP), DX
-
-	// matchLen
-	XORL SI, SI
-	CMPL DX, $0x08
-	JB   matchlen_match4_standalone
-
-matchlen_loopback_standalone:
-	MOVQ (AX)(SI*1), BX
-	XORQ (CX)(SI*1), BX
-	JZ   matchlen_loop_standalone
-
-#ifdef GOAMD64_v3
-	TZCNTQ BX, BX
-#else
-	BSFQ BX, BX
-#endif
-	SHRL $0x03, BX
-	LEAL (SI)(BX*1), SI
-	JMP  gen_match_len_end
-
-matchlen_loop_standalone:
-	LEAL -8(DX), DX
-	LEAL 8(SI), SI
-	CMPL DX, $0x08
-	JAE  matchlen_loopback_standalone
-
-matchlen_match4_standalone:
-	CMPL DX, $0x04
-	JB   matchlen_match2_standalone
-	MOVL (AX)(SI*1), BX
-	CMPL (CX)(SI*1), BX
-	JNE  matchlen_match2_standalone
-	LEAL -4(DX), DX
-	LEAL 4(SI), SI
-
-matchlen_match2_standalone:
-	CMPL DX, $0x02
-	JB   matchlen_match1_standalone
-	MOVW (AX)(SI*1), BX
-	CMPW (CX)(SI*1), BX
-	JNE  matchlen_match1_standalone
-	LEAL -2(DX), DX
-	LEAL 2(SI), SI
-
-matchlen_match1_standalone:
-	CMPL DX, $0x01
-	JB   gen_match_len_end
-	MOVB (AX)(SI*1), BL
-	CMPB (CX)(SI*1), BL
-	JNE  gen_match_len_end
-	INCL SI
-
-gen_match_len_end:
-	MOVQ SI, ret+48(FP)
-	RET
diff --git a/flate/matchlen_generic.go b/flate/matchlen_generic.go
index 8c840f9b4..6149384aa 100644
--- a/flate/matchlen_generic.go
+++ b/flate/matchlen_generic.go
@@ -1,6 +1,3 @@
-//go:build !amd64 || appengine || !gc || noasm
-// +build !amd64 appengine !gc noasm
-
 // Copyright 2019+ Klaus Post. All rights reserved.
 // License information can be found in the LICENSE file.